- provides a duplicate function call suppression mechanism.
- executes and returns the results of the given function, making sure that only one execution is in-flight for a given key at a time. If a duplicate comes in, the duplicate caller waits for the original to complete and receives the same results.
// you have a database with weather information per city and you want to expose this as an API. In some cases you might have multiple users ask for the weather for the same city at the same time.
// => just query the database, and then share the result to all the waiting requests
package weather
type Info struct {
TempC, TempF int // temperature in Celsius and Farenheit
Conditions string // "sunny", "snowing", etc
}
var group singleflight.Group
func City(city string) (*Info, error) {
results, err, _ := group.Do(city, func() (interface{}, error) {
info, err := fetchWeatherFromDB(city) // slow operation
return info, err
})
if err != nil {
return nil, fmt.Errorf("weather.City %s: %w", city, err)
}
return results.(*Info), nil
}
- best described as a
sync.WaitGroup
but where the tasks return errors that are propagated back to the waiter. - useful when you have multiple operations that you want to wait for, but you also want to determine if they all completed successfully.
// you want to lookup the weather for multiple cities at once, and fail if any of the lookups fails.
func Cities(cities ...string) ([]*Info, error) {
var g errgroup.Group
var mu sync.Mutex
res := make([]*Info, len(cities)) // res[i] corresponds to cities[i]
for i, city := range cities {
i, city := i, city // create locals for closure below
g.Go(func() error {
info, err := City(city)
mu.Lock()
res[i] = info
mu.Unlock()
return err
})
}
if err := g.Wait(); err != nil {
return nil, err
}
return res, nil
}
- through the use of semaphores by keeping track of how many tasks are running, and to block until there is room for another task to start.
- to allow up to 10 tasks to run at once, we create a channel with space for 10 items:
semaphore := make(chan struct{}, 10)
- to start a new task, blocking if too many tasks are already running, we simply attempt to send a value on the channel:
semaphore <- struct{}{}
- When a task completes, mark it as such by taking a value out of the channel:
<-semaphore
func Cities(cities ...string) ([]*Info, error) {
var g errgroup.Group
var mu sync.Mutex
res := make([]*Info, len(cities)) // res[i] corresponds to cities[i]
sem := make(chan struct{}, 10)
for i, city := range cities {
i, city := i, city // create locals for closure below
sem <- struct{}{}
g.Go(func() error {
info, err := City(city)
mu.Lock()
res[i] = info
mu.Unlock()
<-sem
return err
})
}
if err := g.Wait(); err != nil {
return nil, err
}
return res, nil
}
- not all tasks are equally expensive => instead of reasoning about the number of tasks we want to run concurrently, we come up with a "cost" for every task and acquire and release that cost from a semaphore.
- golang.org/x/sync/sempahore package provides a weighted semaphore implementation
sem <- struct{}{}
operation is called "Acquire"semaphore.Acquire
method returns an error; that is because it can be used with thecontext
package to abort the operation early.
<-sem
operation is called "Release"
{% code lineNumbers="true" %}
// let's pretend the cost varies with the length of the city name
func Cities(cities ...string) ([]*Info, error) {
ctx := context.TODO() // replace with a real context
var g errgroup.Group
var mu sync.Mutex
res := make([]*Info, len(cities)) // res[i] corresponds to cities[i]
sem := semaphore.NewWeighted(100) // 100 chars processed concurrently
for i, city := range cities {
i, city := i, city // create locals for closure below
cost := int64(len(city))
if err := sem.Acquire(ctx, cost); err != nil {
break
}
g.Go(func() error {
info, err := City(city)
mu.Lock()
res[i] = info
mu.Unlock()
sem.Release(cost)
return err
})
}
if err := g.Wait(); err != nil {
return nil, err
} else if err := ctx.Err(); err != nil {
return nil, err
}
return res, nil
}
{% endcode %}
Resources: